TOPICALIZATION AND OTHER PUZZLES OF GERMAN SYNTAX1

Page created by Ted Jenkins
 
CONTINUE READING
TOPICALIZATION AND OTHER PUZZLES OF GERMAN SYNTAX1
TOPICALIZATION AND OTHER PUZZLES OF
              GERMAN SYNTAX1
                       Hubert Haider, Dept. of Linguistics, Univ. of Stuttgart
                                                    1990

      [in G. Grewendorf & W.Sternefeld eds. Scrambling and Barriers. Amsterdam: Benjamins. p. 93-112]

Abstracts. This paper analyzes the syntactic properties of sentences with a V-projection in clause initial
position, so-called VP-topicalization. The analysis pursues two theoretic claims. First, it will be claimed
that this construction provides an argument for a representational conception of Generative Grammar
(cf. Koster 1987) and against the standard GB-model with derivation by movement. It will be shown
that VP-topicalization defies a derivational analysis. Secondly, this construction provides evidence for
the claim that in German the subject is internal to V-max. In section 2, the relevant syntactic aspects of
the construction are introduced. Section 3 provides arguments that a movement analysis cannot capture
the relevant generalizations. A representational account is presented in section 4, together with an expla-
nation of the syntactic behavior described in section 2. Section 5 discusses some consequences of the
representational account.

1
 I am grateful to Klaus Netter for comments and criticism. It was his observation that pronouns in the Wacker-
nagel position (cf. examples in (12)) behave differently.

                                                      1
1. INTRODUCTION
The two main issues dealt with in this paper are i) arguments for a representational concept of
syntactic representations instead of a derivational one, and ii) arguments in favor of an inclusive
verbal projection for German, i.e. a VP that contains the subject. The data on which the argu-
ments are based is a common construction of German, the topicalization of a verbal projection:
(1) [CP [Einen Blinden geführt]i [C' hatj [X-max ein Einäugiger ei ej]]]
      a blind one guided            has         a one-eyed
     ‘A one-eyed has guided a blind one’
According to a current derivational analysis (cf. den Besten 1983), (1) is the product of two
instances of movement: the finite Verb moves to the C-position, and a phrase moves to the
SPEC-C-position. SPEC-C can host only one phrase, and it must be a maximal projection (cf.
Chomsky 1986). Under a representational analysis, the verbal projection would be generated in
SPEC-C and coindexed with an empty category in the functional base position. It will turn out
that the latter approach is empirically superior to the former.
As to the status of V-max, an SOV language like German or Dutch is in principle open for one
of two types of verbal projections. Either the subject is external to VP, like in English (exclu-
sive VP) or the subject is part of the VP (inclusive VP). There are independent arguments (cf.
Haider 1986, 1989a) that the VP in Dutch is exclusive whereas it is inclusive in German. The
analysis of topicalization will support this claim. Upon first sight, however, it seems to provide
an argument for an exclusive VP:
(2) *[Ein Einäugiger geführt] hat einen Blinden
      a one-eyed guided has a blind
      (a one-eyed has guided a blind)
It is tempting to try to account for the difference between (1) and (2) in terms of the constituent
structure of VP. In (1) a VP seems to be topicalized, while in (2) the subject together with the
verb appears in front, with the object left behind. Since the subject does not form a constituent
with the verb that excludes the object, (2) is illformed, because fronting of the subject and the
verb would encompass the object too. This account is too simple, however, as the examples in
(3) reveal.
(3) a. [Ein Außenseiter gewonnen] hat hier noch nie
    b. [Ein Außenseiter gewonnen] hat es hier noch nie (es = z.B. das Derby)
    c. [Ein Außenseiter gewonnen] scheint hier noch nie zu haben
    d. [Ein Außenseiter gewonnen zu haben] scheint hier noch nie
In each sentence a non-ergative subject is part of the verbal projection. In (3b) we find in addi-
tion an object left behind, and in (3c,d) the subject is part of a verbal projection that contains
even a raising verb. The most exotic kind of evidence comes from split NPs, as illustrated in
(4).
(4) a. [Briefe geschrieben] hat sie mir bis jetzt nur drei traurige
    b. [Außenseiter gewonnen] hat es bis jetzt nur ein einziger

                                                   2
What we find in the initial position is the verb together with the head of an NP, with the rest of
the NP at the end of the sentence. It will turn out that there is no movement process that could
strip an NP of its head and move it to the front, together with the verb. In the following section
I will present evidence for the claim that the topicalized verbal projection is base generated in
its topic position.
2. SOME EMPIRICAL ISSUES
The observations discussed in this section are organized under a particular perspective. I choose
constructions for which the topicalized V-projection displays properties which do not obtain to
its base position. Since these properties would be conserved under movement, however, they
should be found in the base position, too. This will be taken as evidence for an analysis which
assumes the topicalized projection to be base-generated in SPEC-C (see section 4).
2.1 Extraposition
An extraposed clause is usually taken to be adjoined either to VP or S, depending on its argu-
ment status (cf. Reinhart 1983). As shown in (5a,c) and (6a) an extraposed clause may be ad-
joined to the topicalized V-projection.
(5) a. [Fragen, ob wir einverstanden sind] wird er wohl müssen
    b.*daß er wohl fragen, ob wir einverstanden sind, müssen wird
    c. [Schreiben, daß er nicht kommen darf] hat er nicht mehr können
    d.*daß er nicht mehr hat schreiben, daß er nicht kommen darf, können
    e. daß er nicht mehr hat schreiben können, daß er nicht kommen darf
(6) a. [Hunde füttern, die Hunger haben], würde wohl jeder
    b.*daß wohl jeder [Hunde füttern, die Hunger haben] würde
    c. daß wohl jeder [Hunde, die Hunger haben], füttern würde
    d. daß wohl jeder Hunde füttern würde, die Hunger haben
Since the extraposed variant with extraposition cannot occur in the base position (cf. 5b,d; 6b),
extraposition must take place in the topicalized position, i.e. after topicalization. In this case,
however, it cannot be maintained anymore that there are different positions for extraposed rela-
tive clauses in contrast with extraposed argument clauses. Both are adjoined to the same V-
projection (cf. 5a,c and 6a). There is still one more problematic consequence for a derivational
analysis. In a finite clause, the V-projection to which the extraposed clause is adjoined contains
the finite Verb (cf. 6d). Hence we would expect that a clause which is extraposed out of a topi-
calized V-projection is adjoined to the VP of the matrix clause. Thus we would get (7) instead
of (5a).
(7) Fragen wird er wohl müssen, ob wir einverstanden sind
(7) could be generated also if extraposition is applied before topicalization. Since we do not
want to have ordered rules, this is what we would like to find. Since (5a,c) and (6) exist never-
theless, they must be deriveable. Of course, one could assume that the topicalized V-projection
originally contained the finite verb, which afterwards moved to the C-position, leaving a trace
behind:
(8) [Gesungen ei] hati er
                                                 3
Unfortunately, a derivation of the type (8) must not be permitted. If it were possible, we would
get sentences like (9), which are ungrammatical.
(9) a.*[Ihr ein Buch ei] gabi Hans
    b.*[Ein Buch auf ei] schlugi Hans
Hence we end up with a construction for which there is no satisfactory account in terms of VP-
movement, i.e. in derivation-by-movement terms.
2.2 Definiteness effect
It has been noted by Kratzer (1984) that Vn -topicalization is affected by the definiteness effect,
if a subject, ergative or unergative, is part of the projection.
(10)a. Ein Fehler unterlaufen ist ihr noch nie
    b.??Dieser Fehler unterlaufen ist ihr noch nie
    c. daß ihr dieser Fehler noch nie unterlaufen ist
    d. Ein Außenseiter gewonnen hat hier noch nie
    e.??Der Außenseiter gewonnen hat hier noch nie
    f. daß hier noch nie der Außenseiter gewonnen hat
(11) a. There arrived a man from Rio
     b.??There arrived the man from Rio
     c. Es dirigierte Herbert von Karajan
     d. Es gewann die Mannschaft aus Schweden
(11c,d) exemplify that in German there is no definiteness comparable to the one in English or
Dutch in existential or presentative clauses. Again, if it is just a VP that is moved to the front, it
is unclear why this VP should gain new properties.
2.3. Subjects within the topicalized projection
Subjects may appear in the topicalized V-projection precisely under two conditions, which have
to be captured by the derivational account. First, the subject must be non-referential (definite-
ness effect). Secondly, an unergative subject may appear in the fronted V-projection provided
that there is at most one argument left behind which occurs immediately after the finite verb (cf.
12b vs. 12c). It is this property alone, which discriminates ergative and unergative subjects (cf.
12b vs. 13a).
(12) a. Ein Außenseiter gewonnen hat da noch nie
     b.*Ein Außenseiter gewonnen hat da noch nie das Derby
     c. Ein Außenseiter gewonnen hat das da noch nie
     d. Linguisten gespeist haben dort noch nie
     e.*Linguisten gespeist haben dort noch nie Langusten
     f. Linguisten gespeist haben das dort noch nie
     g. Kinder gespielt haben hier noch nie
     h.*Kinder gespielt haben hier noch nie Tempelhüpfen
     i. Kinder gespielt haben das hier noch nie
(13) a. Ein Fehler unterlaufen ist auch schon mal diesem Professor
     b. Ein Fehler unterlaufen ist ihm auch schon mal
                                                  4
c. Ein Tiger entwichen ist doch erst kürzlich diesem Wanderzirkus
    d. Ein Tiger entwichen ist ihm doch erst kürzlich
In (12) I chose three transitive verbs with an optional direct object. If the subject is part of the
topicalized constituent, the object cannot appear in its base position after the adverbials (cf.
12b,e,h). Ergative subjects behave differently (cf. 13a,c). Ergative subjects pattern in this con-
struction like passive subjects, as expected.
(14) a. Ein Job angeboten wurde damals sofort jedem Tagträumer
     b. Ein Job angeboten wurde ihm damals sofort
It will be made clear in the following section that the differences noted above cannot be cap-
tured adequately under a derivational approach employing scrambling.
2.4 Inconsistent structure requirements
If one takes the premises seriously that only maximal projections appear in the SPEC-C posi-
tion and that only one projection can be moved to this position, topicalization structures reflect
inconsistent structure assignment requirements.
(15) a. daß er ihren Argumenten folgen können wird
     b. Ihren Argumenten folgen wird er doch wohl können
     c. folgen können wird er ihren Argumenten doch wohl
According to (15b) we would like to project structure (16a) on (15a), but on the example (15c)
we would like to choose (16b).
(16) a. [[seinen Argumenten folgen] können]
     b. [seinen Argumenten [folgen können]]
One might try to stick to (16a) and derive (15c) by string-vacuous Chomsky-adjunction of the
object to its mother VP and move the emptied VP. This move would still leave it mysterious,
however, how (17a) and (17b) could be tackled.
(17) a. [Ein Außenseiter gewinnen] hätte hier wohl kaum können
     b. [Gewinnen können] hätte ein Außenseiter hier wohl kaum
(17a) requires a structure in which the modal takes an S-complement, which is fronted. In order
to derive (17b), the subject must be Chomsky-adjoined again to its mother constituent, the S,
such that the emptied constituent contains only the modal. This cannot be the case, however,
since S would contain also the adverbials and the finite verb, too. It is shown above that this
would give rise to a construction like (9) and therefore has to be abandoned. This leaves (17)
without a derivational source.
2.5 Topicalization of non-constituents
As pointed out in section 1, split NPs pose a serious problem for a derivational analysis, if they
appear within a verbal projection. For convenience, the examples (4) are repeated under (18).
(18) a. [Briefe geschrieben] hat sie mir bis jetzt nur drei traurige
     b. [Außenseiter gewonnen] hat es bis jetzt nur ein einziger
(18) is the VP-topicalization variant of (19).

                                                  5
(19) a. Briefe hat sie mir bis jetzt nur drei traurige geschrieben
     b. Außenseiter hat es bis jetzt nur ein einziger gewonnen
The characteristics of constructions of the type (19) are the following. The element in SPEC-C
must be the head of the split NP (cf. 20a). The NP must be indefinite (cf. (20b). The NP con-
tains a gap (cf. 20c). The relation between the two parts obeys movement constraints (cf. 21).2
(20) a.*Nur drei traurige hat sie mir bis jetzt Briefe geschrieben
     b.*Briefe hat sie mir bis jetzt nur die drei traurigen geschrieben
     c.*Briefe hat sie mir bis jetzt nur drei traurige Episteln geschrieben
(21) a. Bücher sagte man mir, habe er nur politische geschrieben
     b.*Bücher sagte man mir, wer nur politische geschrieben habe
     c.*Bücher nannte sie mir einen Mann, der nur politische geschrieben hat
The fact that (19) does not permit a split variant in the base position, bars a movement account
for (18).
(22) a.*daß sie nur drei traurige bis jetzt Briefe geschrieben hat
     b.*daß es nur ein einziger bis jetzt Außenseiter gewonnen hat
There is no V-projection consisting of the head of an NP and the verb.
2.6 Scope asymmetries3
(23) shows, that scope properties are subject to reconstruction.
(23) a. daß Max jemandem kein Buch verkaufen darf
     b. Max darfi [jemanden kein Buch verkaufen ei]
Both in (23a) and in (23b) the negation may have scope over the modal. This can be accounted
for by reconstruction: Since the trace of the modal is in the scope of the negation this carries
over to the antecedent. That the middle field, i.e. the constituent in brackets in (23b) constrains
the scope domain can be seen from (24):
(24) a. daß jedem Lehrer ein Schüler gefallen möchte                   (narrow, wide)
     b. daß ein Schüler jedem Lehrer gefallen möchte                   (narrow, wide)
     c. [Jedem Schüler gefallen] möchte ein Lehrer                     (wide)
To account for (24), we have to assume both that the middle filed restricts the scope domain (as
in 23), to keep the existential quantifier in a narrow scope and we have to assume that the topi-
calized constituent does not reconstruct. Otherwise we would expect the same scope properties
as in (24a,b). If (24c) is derived by movement we expect reconstruction, however.
2.7 Independent evidence for base generated V-projections
Left-dislocation provides evidence that there are base-generated V-projections in non-base posi-
tions on the one hand, and that there are differences between left-dislocation and topicalization
2
  Henk van Riemsdijk discussed these constraints in his talk given at the Workshop on Dialectology in Venice,
April 1, 1987. In V-projection structures of the type (18) the remnant of the NP tends to be clause final, i.e. imme-
diately adjacent to the gap the V-projection in SPEC-C is coindexed with. This information I owe to Gosse Bou-
ma, personal communication.
3
  These data I owe to T.Höhle, who discussed them in a talk given at the GGS-workshop Konstanz, November
1986.
                                                         6
on the other hand that provide insight into the nature of the dependency between the left periph-
eral position, be it SPEC-C or the LD-position, and its functionally equivalent clause internal
position. Since the dependency is different, it is unlikely that one can be derived from the other,
as suggested e.g. by Koster (1978). The following sample of contexts is sufficient to dismiss the
claim that topicalization is derived from left-dislocation simply by replacing the pronoun in the
SPEC-C Position by an empty operator.
(25)a. anaphor:
     Sich (*den) mag er am liebsten
     [AP stolz auf sich] (das) war er schon immer
  b.quantifier
     Jeder (*der) weiß das
  c. indefinite pronoun
     Jemand (*der) hat sich geirrt
  d.sentential adverb
     Vermutlich (*das) weiß er es
  e. predicatives4
     Stumm (*das) lächelt Mona Lisa
  f. NP-subconstituents
     Linguisten (*die) kenne ich nur kluge
  g. Extraposition out of a topicalized V-projection
     Rosen gezüchtet, (*das) hat er die prämiert wurden
  h.V-projection with subcategorized elements
     Gewartet (*das) hat er auf sie nicht
  i. Split NPs in V-projections
     Briefe geschrieben (*das) hat sie drei traurige
  j. V-Projections that contain a non-ergative subject
     Ein Außenseiter gewonnen (*das) hat hier noch nie
These differences call for an explanation in terms of the syntactic relation that holds between
the non-base position and the functional base position. For topicalization this relation is an an-
tecedent-empty category relation, for LD, however, it is a relation between a phrase and a pro-
nominal element it is coindexed with. Whatever account is given for this difference, this does
not concern the fact that in LD-constructions we find verbal projections generated in non-base
positions:
(26) a. [Ihr Blumen geschenkt] (das) hat er noch nie
     b. [Ein Buch lesen] (das) würde er nie
     c. Gelogen (das) hat er noch nie
     d. Geregnet (das) hat es schon lange nicht mehr
3. SCRAMBLING - AN INADEQUATE SOLUTION
There are some recent attempts (cf. den Besten & Webelhut 1990) to revive Ross's idea how to
handle free word order, namely by employing the concept of scrambling. Scrambling is inter-
preted as the result of free adjunction to VP, a concept introduced by Chomsky (1986) in the
4
    Marga Reis, personal communication.
                                                 7
framework of "Barriers". According to a scrambling-based approach the topicalized V-
projection in (27) is V-max in all cases:
(27) a. [ ei ej geöffnet] hat sie ihmi die Augenj
     b. [ ei die Augen geöffnet] hat sie ihm ei
     c. [ihm die Augen geöffnet]          hat sie
In order to derive (27), the objects must be moved out of the VP. Adjunction of the objects to
VP should leave a preposable empty VP. The result is (28) as a base structure for (27a).
(28) [CP [C´ [VP sie [VP ihmi [VP die Augenj [VP ei ej geöffnet ]]] hat]]]
It is easy to demonstrate that scrambling-by-adjunction is both too strong and too weak a con-
cept. It is too strong because it overgenerates and it is too weak because there are topicalization
structures which cannot be derived by means of scrambling. Let us start with the case of over-
generation:
A scrambling approach neither captures the definiteness effect (cf. 2.2) nor the subject-object
dependency discussed in section 2.3. (29) shows that objects may appear in front of a definite
subject. This means that the object is adjoined to S, which turns the basic S into a candidate for
topicalization:
(29) a. daß den Mann die Kinder nicht gestört haben
     b.*[Die Kinder gestört] haben den Mann nicht
The fact that a subject may appear in the topicalized projection only if the remaining object - in
case there is one - appears in front of the middle field does not follow from the way how
scrambling works. For convenience, I repeat an example illustrating the relevant contrast:
(30) a. Ein Außenseiter gewonnen hat dieses Jahr noch nie
     b.*Ein Außenseiter gewonnen hat dieses Jahr noch nie das Derby
     c. Ein Außenseiter gewonnen hat es dieses Jahr noch nie
For (30b) there exists a perfectly well-formed scrambled base-variant as a source for topicaliza-
tion:
(31) a. daß ein Außenseiter dieses Jahr noch nie das Derby gewonnen hat
     b. daß das Derby ein Außenseiter dieses Jahr noch nie gewonnen hat
     c. daß dieses Jahr noch nie das Derby ein Außenseiter gewonnen hat
First the object is adjoined to S, which yields (31b). Then the adverbials are adjoined, which
yields (31c). Now the emptied S-constituent is topicalized, which yields the ungrammatical
(30b).
Scrambling is to weak for the derivation of the split-NP cases. For (32) there is no source for
topicalization because splitting an NP by scrambling is ungrammatical, cf. (19a), (18a) and
(22a), repeated under (32a), (32b) and (32c) respectively.
(32) a. Briefe hat sie mir bis jetzt nur drei traurige geschrieben
     b. [Briefe geschrieben] hat sie mir bis jetzt nur drei traurige
     c.*daß sie mir nur drei traurige bis jetzt Briefe geschrieben hat

                                                    8
(32c) shows that the NP cannot be split within the middle field. A sentence like (33) would re-
quire a base structure like (32c).
(33) [Briefe geschrieben] hat sie mir nur drei traurige bis jetzt
Scrambling is too weak, moreover, because it is completely unclear under that hypothesis why
the topicalized V-Projection is an extraposition site (cf. sect. 2.1). It is too weak also because it
overgenerates, if V-projections are topicalized that contain the finite verb.5
(34)*[Ihr ein Buch ei] schenktei er.
The ungrammaticality of (34) cannot be attributed to the fact that the projection contains the
trace of the finite verb, because exactly this trace occurs in the case of what would be an S-
topicalization:
(35) [Ein Außenseiter gewonnen ei] hati hier noch nie
Since the topicalized constituent contains the subject, it must be topicalization of S. S, however,
is IP and hence contains the finite verb. Scrambling is too strong again, because it rules out per-
fectly well-formed structures in the case of coordination:
(36) weil esi entweder [niemand ei bemerkte] oder [Maria alle bestochen hat]
Under a scrambling analysis (36) violates the Across-the-board constraint, which forbids con-
joining a phrase containing a gap with another phrase that does not contain a corresponding
gap. Scrambling leaves a gap in the base position. Finally it should be noted that a scrambling
approach towards topicalization frequently involves vacuous movement. In (37) the object is
string-vacuously adjoined to VP in order to turn the topicalized element into a VP. Chomsky
(1986) provided arguments against vacuous movement, which can be strengthened (cf. Haider
1989b) to a complete ban.
(37) [ei geholfen]j hat [sie [VP ihri [VP e]j
In the following section I will try to show that all these problems do not arise if we give up the
idea that topicalization is the result of a movement process.
4. TOWARDS A REPRESENTATIONAL ACCOUNT OF TOPICALIZATION
Let us suppose that the V-projection in the SPEC-C position is base-generated in this very posi-
tion just like the V-projection is base-generated in LD-constructions. These two constructions
differ with respect to the relation they enter with the functional base position. For topicalization
it is an antecedent - gap relation, for LD it is an antecedent - pronominal relation. The pronomi-
nal in (26) acts as a predicate pronominal whose interpretation is determined by the predicate it
is coindexed with. If there is no predicate, the pronominal is interpreted deictically (cf. 26 and
38).
(38) a. Das hat er noch nie
     b. Das würde er nie
     c. Das hat er noch nie
     d. Das hat es schon lange nicht mehr

5
    This has been noted by Jindrich Toman, in the discussion period.
                                                          9
Semantically, the verbal projection in an LD-construction is interpreted as an autonomous pred-
icate, whose interpretation determines the interpretation of the pronominal it is coindexed with.
The analogous analysis for topicalization entails that the topicalized V-projection is interpreted
as an autonomous predicate, too, i.e. as a maximal projection of V. In LF this V-projection will
receive a semantic interpretation in its SPEC-C position. Since the topicalized phrase enters
into an antecedent - gap relation by virtue of its occupying the SPEC-C position, the predicate
in SPEC-C position will be related to the matrix predicate as if the topicalized predicate were a
single verb:
(39) Vi hatj er nicht [VC [ ei ] ej]]
The only well-formedness requirements for the verbal projection in (39) are a syntactic one and
a functional-semantic one. It must be a maximal projection and it must provide a Θ-role for the
subject in the middle field. Hence any of the following instances of Vn will give a grammatical
sentence in the context of (39):
(40) a. getanzt
     b. mir geholfen
     c. mich absichtlich geohrfeigt
     d. ihr heimlich einen Kuß gegeben
     e. ihr einen Kuß auf die Wange gedrückt, daß es schnalzte
From a semantic point of view, these are intransitive predicates. Some are basically intransitive
(40a), some are intransitive due to saturation of all argument slots except one. Hence they fulfill
the functional requirement of the gap the predicate in (39) is coindexed with. It is the gap of an
intransitive verb, a gap that would arise if we inserted a verb like (40a) and moved it to the
SPEC-C position. We could not insert the other predicates, however, since it is the slot of a V°
category in the verbal cluster. The verbal cluster does not contain phrases. Hence these phrases
do not have a derivational source. Let us investigate now how this hypothesis allows handling
the recalcitrant problems discussed in section 2 and 3.
4.1 Extraposition
In German, and presumably in all other languages as well, Vmax is the adjunction site for extra-
position, both for clausal arguments and non-arguments. Claims that different binding proper-
ties require different adjunction sites (cf. Reinhart 1983) are ill-founded. It is not the adjunction
position that matters but the argument status. Principle C violations do not arise, despite of a c-
commanding, preceding, coindexed NP, if the noun is contained in a non-argument clause. This
is easy to see in German, where adverbials and likewise adverbial clauses appear between an
object and the verb, i.e. VP-internal:
(41) a.   Der Vorsitzende hat ihri [obwohl Mariei heftig protestierte] das Wort entzogen
     b.   daß eri [als man Hansi davon berichtete] kollabierte, wundert mich nicht
     c.   Man hat ihmi mehr Geld als Hansi sich erwartet hatte angeboten
     d.   Man hat ihri das Haus, das Mariai bekanntlich geerbt hat, streitig gemacht
Non-argument clauses are opaque for principle C. Viewed from this perspective, it is not sur-
prising that relative clauses are adjoined to the same V-projection in the topicalized position as
object clauses; they are adjoined to the same projection in the base position as well. The V-
                                                 10
projection in SPEC-C position is a maximal V-projection and hence an adjunction site for ex-
traposition.
4.2 Definiteness Effect
A definiteness effect is found whenever a subject does not appear in its canonical position. This
is easy to check for English. In a there-construction, the thematic subject appears VP-internally,
in a non-canonical subject-position. In German, the canonical subject position is in the middle
field and not in a V-Projection in SPEC-C. Hence the topicalized V-projection displays a defi-
niteness effect for the subject. The fact that there is no definiteness effect in the middle field
simply shows that any position in the middle field qualifies as canonical position for a subject.
This is to be expected if German has an inclusive VP.
4.3 Subjects within the topicalized VP
If the topicalized VP contains a subject, it must be non-referential, i.e. indefinite and, if it is
non-ergative, its co-argument must not occur inside the middle-field. This follows from a clo-
sure property induced by non-ergative subjects.
(42) The verbal projection that contains the non-ergative subject is closed for the projection
     principle
What (42) amounts to is that a Θ-role cannot be passed on to an argument outside a VP that
contains a non-ergative subject. Hence in the example (43) the object remains Θ-less and vio-
lates the Θ-criterion.
(43) *Ein Außenseiter gewonnen hat da noch nie das Derby (cf. 12b)
What is the difference between (43) and (44) that allows maintaining (42), although the object
is not in the same V-projection?
(44) Ein Außenseiter gewonnen hat das da noch nie (cf. 12c)
There is one possibility, how the object might receive a Θ-role, namely by means of a chain:
(45) [Ein Außenseiter ei gewonnen] hat dasi da noch nie
What we have to explain now is why a chain is possible in (45) but not in (43). The answer is
simple. In (44) the pronoun occupies the so-called Wackernagel-position, a position that imme-
diately follows C. This is an adjunction position of the middle field. It is easy to show, that the
c-command requirement for the antecedent of the gap can only be met in that position and not
in the base position, which will explain the difference between (43) and (44). All we have to do
is to adapt the definition of dominance to adjunction structures. According to Chomsky (1986),
adjunction produces segments of categories.
(46) [XP Y [XP Z ]]              (adjunction of Y to XP)
In (46), XP consists of two segments. Given that a category C includes an element E, if every
segment of C dominates E, Z is included by XP, but Y is not. Now it is easy to adjust domi-
nance to adjunction:
(47) a. A category C c-dominates X, iff C includes X
     b. A c-commands B iff

                                                11
a. A does not c-dominate B
           and
        b. every maximal projection that c-dominates A c-dominates B
(47) is a conservative extension of the original definitions, since in non-adjunction structures
inclusion coincides with domination?
The difference between (43) and (44) follows immediately from (47): It is only in the
Wackernagel-position, not in the base position, that an object can c-command an empty catego-
ry in SPEC-C, because for the element in that position CP is the only maximal projection that
includes it and CP dominates the gap. The difference between this account and the scrambling
account is clear. Under the latter hypothesis, both the adverbial and the object would be in ad-
joined position and hence the object should be able to c-command, which is obviously not the
case, as (48) illustrates:6
(48)*Ein Außenseiter gewonnen hat da noch nie das Derby (cf. 12b)
4.4 Inconsistent structure assignments
The problem of inconsistent structure requirements arises only under the movement analysis. If
a V-projection is base-generated in SPEC-C, it can be any V-projection, provided it meets the
functional requirements imposed on it by the kind of arguments that appear in the middle field.
(49) a. Ihren Argumenten folgen wird er doch wohl können (cf. 15b)
     b. Folgen können wird er ihren Argumenten doch wohl (cf. 15c)
In (49a) the topicalized V-projection can be replaced by any intransitive predicate, i.e. anything
that is functionally equivalent, e.g. by the intransitive variant of ‘folgen’, as in (50a). In (49b)
the functional properties are not affected by the modal, hence its omission would lead to (50b).
(50) a. Folgen wird er doch wohl können
     b. Folgen wird er ihren Argumenten doch wohl
Since this account is not bound to the assumption that the topicalized V-projection must corre-
spond to an isomorphic structure in the base position before movement, the structural paradox
cannot arise.
4.5 Topicalization of non-constituents
The definiteness affect observed with this construction indicates that the head noun in the V-
projection is interpreted as a predicate. Evidence to this end comes from (51).
(51) a. [Briefe geschrieben] hat sie nie welche
     b.*daß sie nie welche Briefe geschrieben hat
     c. daß sie nie welche geschrieben hat

6
  Günther Grewendorf, p.c., noted that (48) can be improved if the object precedes the adverbial and carries focus
stress:
  i) Ein Außenseiter gewonnen hat das DERBY hier noch nie
The difference between i) and (48) follows immediately if it is recognized that the focus position for non-
pronominal NPs coincides with the Wackernagel-position. Then the analysis given for (45) carries over to i) as
well.
                                                       12
The indefinite pronoun ’welche’ represents an NP. Hence it will receive the Θ-role from the
verb. But then no Θ-role is left for the noun ’Briefe’ in (51a), which cannot be part of the ob-
ject-NP, as (51b) shows. If it does not receive a é-role it cannot be an argument but only a pred-
icate. This seems to match the interpretation of (51a). ’Briefe’ specifies the denotation of the
pronoun just like in (52).
(52) Das sind Briefe
It seems that the role of ’Briefe’ in (51a) is the same as in (53).
(53) [Briefe geschrieben] hat sie mir bis jetzt nur drei traurige
The noun restricts the denotation of the element the Θ-role is assigned to but it does not bear it.
Thus it does not close the V-projection even if it is in relation with a non-ergative subject, as in
(54). If the head of the subject NP in the V-projection would close it, the remnant of the NP
could not get its Θ-role.7
(54) [Außenseiter gewonnen] hat es bis jetzt nur ein einziger
The fact that the noun serves as a predicate correlates with the fact that it cannot appear without
a concomitant NP that it agrees with. It is licensed only under predication (cf. 55)
(55) a.*[Außenseiter gewonnen] hat es bis jetzt nicht8
     b.*[Außenseiter gewonnen] das hat es bis jetzt nur ein einziger
In (55a) there is no phrase the noun is coindexed with, and in (55b) the phrase it should be
coindexed with is not accessible.
4.6 Scope asymmetries
If the topicalized VP is base-generated in SPEC-C the sope properties become transparent. We
can explain why (56c) does not have the same sope properties as (56a,b).
(56) a. daß Max jemandem kein Buch verkaufen darf                          (NEG: narrow or wide)
     b. Max darf jemandem kein Buch verkaufen                              (NEG: narrow or wide)
     c. Jemandem kein Buch verkaufen darf Max                              (NEG: narrow)
Scope is constrained by the VP. In (56a) and (56b) the negation and the modal share the same
VP. In (56c) the negation is contained in a VP different from the VP that contains the modal.
Hence the negation does not have scope over the modal outside its V-projection.
4.7. Crossing constraint
The scrambling analysis suffers from an explanatory deficiency. It is unclear why there shows
up a crossing effect in some cases (cf. 58) and not in others (cf. 57)
(57) a. [ei ej geöffnet]k hat sie ihmi die Augenj ek
     b. [Ein Außenseiter gewonnen ei] hati hier noch nie

7
  That the V-projection is not closed at all, neither for the Θ-role of the subject nor for that of an object is illustrat-
ed by (i.): [Außenseiter gewonnen] haben bis jetzt ein Derby nur wenige
8
  It is relevant here, that the NP is singular in number. So we know that Außenseiter is a noun and not an NP, since
the NP requires an article. This is different for the plural. In i) Außenseiter is an NP without article, hence i) is
wellformed.
 i) Außenseiter gewonnen haben bis jetz noch nie
                                                            13
(58) a.*[Ihr ein Buch ei] schenktei er
     b.*[ei mit]PP haben sie dai nicht gerechnet
     c. Dai haben sie nicht [ei mit]PP gerechnet
Under the base generation hypothesis, (58a,b) but not (57a,b) turn out to be violations of a
crossing constraint:
(60) a. [Geöffnet]k hati sie ihm die Augen ek ei (cf. (57a)
     b. [Ein Außenseiter gewonnen]j hati hier noch nie ej ei (cf.57b)
     c. [Ihr ein Buch ei]j schenktei er ej ei
     d. [ei mit] haben sie dai ej nicht gerechnet
In (60c,d), but not in (60a,b) we observe a particular type of crossing dependences, which are
ruled out on independent grounds (cf. Pesetzky 1982). In (60a,b) the two dependencies are dif-
ferent in type, a head-movement dependency and a phrase movement dependency. Hence cross-
ing is irrelevant. In (60c,d), however, we find crossing dependencies of the same type, which is
ruled out.
5. SOME REQUIREMENTS FOR AND CONSEQUENCES OF A REPRESENTATIONAL ACCOUNT
The claim that V-Projections are base-generated in SPEC-C requires some clarifications as to
how case- and Θ-assignment applies. The fact that there are VP internal nominatives under any
analysis in a Government & Binding framework (cf. den Besten 1985) irrespective of the par-
ticular assumptions about the canonical position of the subject should suffice to indicate that
VP-internal NPs are accessible for nominative assignment for reasons independent from our
present concern. Therefore I will not elaborate on that matter.
What is unique for the present claim, however, is that a clause may have two independent V-
projections and that the verb in one projection may assign Θ-roles to arguments in the other
projection. The crucial concept for handling this situation is the syntactic role of auxiliaries. In
Haider/Rindler-Schjeve (1987) and in Haider (1986), I tried to demonstrate that the finite verb
is the syntactic main verb of the clause. Auxiliaries can act syntactically as main verbs by virtue
of the transfer of the argument structure from the verb they govern in the verbal complex to the
auxiliary. This mechanism applies without further complications to the topicalized V-
projection, since they are coindexed with a verbal empty category in the base position that is c-
commanded by the auxiliary. Under this perspective there is no direct transfer of the Θ-roles
from the verb in the topicalized projection to the arguments in the middle field just like there is
no direct assignment of Θ-roles from the main verb to the arguments in the middle field. In both
cases it is the auxiliary which mediates the Θ-role assignment and case assignment.
One of the consequences of the representational account is that German has an inclusive VP,
i.e. a V-projection that contains the subject. Since it is beyond dispute that Dutch does not have
an inclusive VP, we predict specific differences for the topicalization structures. Topicalized V-
projections that contain unergative subjects cannot occur in Dutch since non-ergative subjects
cannot occur in V-Projections in Dutch. The following examples confirm this claim:
(61) a. *Een buitenstaander gewonnen heeft daar nog nooit (cf.12a)
     b. *Linguisten gegeten hebben daar nog nooit (cf.12b)
     c. *Kinderen gespeeld hebben hier nog nooit (cf. 12c)
                                                   14
This difference falls in line with a lot of other systematic differences between subjects in Ger-
man and in Dutch (cf. Haider 1986, 1989a) that receive an explanation in terms of the different
status of V-max in the respective languages.

6. CONCLUSION
The aim of this paper is twofold. From the empirical point of view, it presents an analysis that
covers the properties of topicalized V-projections. Topicalized V-projections are base-generated
in SPEC-C. From the theoretical point of view, it is argued that the analysis of this construc-
tions provides arguments for i) a representational view instead of a derivational one, and ii) an
inclusive VP in German, i.e. a V-max that contains the subject, unlike the English or Dutch VP.
A scrambling analysis is dismissed on empirical grounds.

BIBLIOGRAPHY
BESTEN, H. DEN 1983. On the interaction of root transformations and lexical deletive rules.
  In W. Abraham (Ed.), On the Formal Syntax of the Westgermania, Amsterdam: John Ben-
  jamins. (p. 47–131)
BESTEN,H. DEN & G.WEBELHUTH 1990. (this volume). Stranding. In G. Grewendorf & W.
  Sternefeld eds. Scrambling and Barriers. Amsterdam: Benjamins. (p.77-92)
CHOMSKY,N. 1986. Barriers. Cambridge,Mass.: MIT-Press
HAIDER,H. 1986. Deutsche Syntax, generativ - Parameter der deutschen Syntax. Habilita-
  tionsschrift. Universität Wien.
HAIDER,H. 1989a. Θ-tracking systems - evidence from German. In: Maracz,L. & P. Muysken
  eds. Configurationality: the typology of asymmetries. Dordrecht: Foris (p. 185-206)
HAIDER,H. 1989b. Matching projections. In: A.Cardinaletti & G.Cinque & G.Giusti eds.
  Constituent Structure. Dordrecht: Foris (p.101-121)
HAIDER,H. & RINDLER-SCHJERVE,R. 1987. The parameter of auxiliary selection. Linguis-
  tics 25: 1029-1055
KOSTER,J. 1978. Locality principles in syntax. Dordrecht:Foris.
KOSTER,J. 1987. Domains and Dynasties. Dordrecht: Forsis
KRATZER,A. 1984. On deriving differences between German and English. Ms. Technische
  Universität Berlin
PESETZKY,D. 1982. Paths and categories. Unpubl. MIT-Dissertation
REINHART,T. 1983. Anaphora and semantic interpretation. London: Croom Helm

                                               15
You can also read